fix(pycc_types): reject unrenderable instance and protocol string conversion with C0001 (#977) - #985

...nversion with C0001 (#977)
`print(<instance>)` and `f"{<instance>}"` of a class instance that neither
MIR rewrite renders passed `pycc check` and panicked in `pycc build`, and
several builtin-exception-name shadowing shapes passed `check` and aborted
at runtime. A new `pycc_types::string_conversion` predicate, called at the
f-string interpolation and `print` argument sites, now rejects every such
shape with `C0001`: the 25 builtin exception names are decided by
provenance (seeded synthetic class, or unseeded flat seven), every other
name by dataclass shape, and a protocol-typed value with its own message.
Comment-only corrections in `pycc_mir`/`pycc_codegen` state the real gate.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

...ecs (#977)
Add ADR D-237 (predicate table, the two divergences from MIR's name-first
resolution, alternatives, the newly rejected dataclass-under-exception-name
shape), regenerate the decisions index, amend the dataclass rendering
sentence in TYPE_SYSTEM.md and ROADMAP.md, qualify the 2026年08月01日 tuple
follow-up clause, add the v0.4 #977 row, describe the pycc_types C0001
emission and both message shapes in DIAGNOSTICS.md and the C0001 explain
prose, write session snapshot 2026年09月06日-13, and add the PR #971
eighteen-round retrospective entry.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

... close a helper coverage gap (#977)
Review round 1: add public-CLI rejections for a non-dataclass subclass of a
dataclass, a plain PEP 695 generic instance, and `print(self)` inside a
method, so every row of D-237's repro table is pinned; replace the
`assert_accepted` helper's never-taken panic branch with `expect` so the
100% region gate holds.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

#977)

#977)

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

...zation; rename help for builtin-named classes
Codex round 1 on PR #985.
P1: `instantiate_generic_class_methods` built every specialization with
`is_dataclass: false` and an empty `dataclass_fields`, so `check` accepted
`print(Box[int](1))` against the origin `@dataclass Box` while `build`
re-inferred the rewritten call against `0gen_Box__T_int` and the D-237 gate
rejected it (before the gate, MIR skipped the `__repr__` rewrite for the
same reason and codegen panicked). The specialization now carries the
origin's `is_dataclass` and its `dataclass_fields` substituted like `attrs`;
`methods` already carried the mangled synthesized `__repr__`, so the
specialization renders as `Box(n=1)`.
P2: the help recommended `@dataclass` for a class under a builtin
exception name, which the predicate rejects by name regardless; such a
class now gets a rename remedy instead.
Tests: four new unit tests (help variants, specialization metadata via
`check_and_resolve` on both a parsed and a hand-built module) and two
public-CLI tests (generic dataclass check/run consistency; rename help via
`--error-format json` on a `print(self)` shape, since a call to the
shadowing class is reported by the solver's own message under D-220).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

... PR #985)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

...rsion sites
Codex round 2 on PR #985. `pycc_mir` erases `cast(Base, d)` to `d` and its
`__repr__` rewrite keys on the erased value's class, so an upcast from a
plain subclass to a `@dataclass` base passed the type-only gate as `Base`
and panicked in codegen's `to_str` as a `Derived`. The new
`reject_unrenderable_expr` gates the expression's type and then the value
under every erased `cast` (nested included; a user `def cast` is not
looked through), at both the `print` and f-string sites.
Tests: five unit tests (print, f-string, nested cast, cast to the
dataclass's own type accepted, user `def cast` not looked through) and a
public-CLI rejection for the reported shape.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

...ound 2 and the bound-cast follow-up
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

... string-conversion descent
`check_cast` has already inferred the same value successfully by the time
`reject_unrenderable_expr` descends into an erased `cast`, so the `?` on that
second inference was an unreachable branch and left one uncovered region under
the D-014 100%/100% gate. State the invariant with `expect` instead.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

... merging main
`origin/main` gained #986, which pushed `docs/ROADMAP.md` past the 168960-byte
per-resource llms.txt budget `scripts/check-site.sh` enforces once this branch's
own #977 paragraph merged alongside it. Condense the #977 landing paragraph and
the sentence duplicating it inside the #378 paragraph, and re-anchor the session
file's resume steps on the merged state.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>